Interesting assumption about a future with frontier-level open-weight models. Do you see Wolffish primarily as an agent framework, or as a "personal operating system" for AI workflows?
The invite only access puts the owness on the operator of the platform instance to be responsible for what's shared there.
It's designed for people that already know each other who need a private space, not a large platform for anybody.
There is also some built in protection for when someone new is invited to the server. Each member needs to allow that new member access. The idea is that I, the server operator may have welcomed someone to the community, but a server member may not feel the same way, so their content is not shared with that new member until they decide too.
That's an interesting model. It sounds closer to a private club than a traditional social network, where trust is negotiated by the community rather than just the administrator.
Out of curiosity, how many generated boards get rejected before finding a guess-free puzzle at the higher difficulty levels? The 3,000-attempt limit caught my attention.
By far the most irritating limitation is code size. This is a fantasy console, but the Lua VM is compressed into some sort of bytecode stream with a hard limit on the number of "tokens" (basically whole words) in the code. This leads to some oddities, like "e.pos.x" being 5 tokens, while "e.pos_x" is only 3.
My usual platform is NES, and I've got a pretty good handle on how to optimize assembly language to squeeze code down to size, but I haven't really developed that intuition for pico8's Lua implementation. The final version of the game uses 7989/8192 tokens altogether. It just barely squeaks by!
There are some minification tools I could have run, and I think with hindsight there are some architectural changes and code habits I could develop to make this slightly less of an issue. In practice, I'm now learning picotron instead. (It's got actual widescreen! Much more practical for modern displays, I think.)
